A study published last year in The Lancet concluded that “police killings of unarmed black Americans have adverse effects on mental health among black American adults in the general population.” They used Mapping Police Violence data, wherein individuals are coded as “unarmed” if they had a toy/replica gun or if they were “alleged to be armed by the police, but multiple independent witnesses maintain the person was unarmed.”
These are hugely important caveats. Toy/replica guns can be nearly indistinguishable from actual guns, especially in a matter of split seconds. This is why similar databases compiled by The Washington Post and The Guardian are careful to label such incidents as involving toy/replica guns, rather than no weapon at all. Likewise, it is debatable whether we should dismiss police accounts of an incident simply because two or more witnesses disagree.
One Saturday last summer, my colleague, James Lozada, and I decided to take a closer look at 110 fatal officer-involved shootings (OIS) that occurred from 2015 to 2017 involving a black citizen who MPV coded as “unarmed.” We focused specifically on OIS deaths because (1) these account for the vast majority of citizen fatalities at the hands of police, and (2) WAPO’s data do not include deaths by other means. Among these 110 OIS, 36 involved individuals who were not unarmed according to WAPO and The Guardian. Most had a toy/replica firearm, but four had an actual firearm, one had a knife, and three were undetermined according to WAPO:
The Lancent published our written correspondence earlier this year, though unfortunately, we were limited to 250 words. Our original submission was approximately 1,000 words and included tables that listed the discrepancies between MPV and WAPO/The Guardian.
Our main takeaway was that the study’s conclusions may have been overstated, since MPV technically miscodes so many incidents each year. In fairness, the difference between an OIS involving someone who was “unarmed” and “holding/pointing something that looks like, but is not, an actual gun” may be irrelevant in terms of how the community views the incident after the fact. But the authors could have easily recoded these incidents and run sensitivity models to see if their substantive findings changed. At a minimum, researchers must be clear about these little, but critical, details.
